Amir H. Mardan

About Me

A Computational Seismologist & Machine Learning Practitioner

Hello! I'm Amir. I have a PhD in Earth Science and I study Geophysics (seismology). For my master's degree, I used machine learning to detect channels and faults in seismic data. In addition to my thesis work, I used clustering techniques for facies analysis in multiple oilfields. During my PhD, I worked on full-waveform inversion (FWI) to apply it as a time-lapse tool (TL-FWI) to monitor changes in the subsurface that could occur as a result of extraction from, or injection into, reservoirs. TL-FWI has an important role in monitoring the saturation of injected CO₂ for carbon capture and storage (CCS). Currently, I'm a postdoctoral researcher developing a processing pipeline using artificial intelligence (AI) to process near-surface seismic data.

GitHub Linkedin CV Publications PyFWI

Research Interests

Currently, I'm focusing on full-waveform inversion (FWI) and its applications for monitoring the subsurface. Having experience in using machine learning techniques for facies analysis and feature detection in seismic data, I'm also interested in using deep learning for velocity analysis and fluid monitoring in the subsurface. Here are short summaries of the projects that I've worked on.

Deep learning techniques for estimating petrophysical properties

Comprehensive knowledge about the petrophysical properties of the subsurface is crucial for both oil E&P and CO₂ sequestration. I have developed tools to estimate petrophysical properties using a physics-informed neural network.

Automatic first-break picking using deep learning

Manual first-break picking is a time-consuming and labor-intensive task. This processing step is also subjective depending on who does the processing. At Geostack, we have developed tools to perform automatic first-break picking using deep learning.

Automatic dispersion curve picking using deep learning

In a velocity dispersion panel, dispersion curves should be picked and inverted to estimate an image of the subsurface. We have developed a Python package that can be used to pick the fundamental mode automatically using deep learning.

Deep learning techniques for velocity analysis

Estimating an accurate velocity model of the subsurface is critical in seismic studies. Although FWI has been employed for this purpose, this approach suffers from numerous technical problems. In my current project, I've been using deep learning methods to obtain an appropriate velocity model.

Reservoir monitoring using time-lapse FWI

FWI is a powerful technique to recover high-resolution images of the subsurface. Time-lapse FWI can reveal highly valuable information that is significantly important in the petroleum industry for reservoir management. This method can also be employed to monitor the saturation of CO₂ for the purpose of carbon capture and sequestration.

Seismic data interpretation using machine learning techniques

Machine learning techniques can be used to improve our understanding of seismic data. Supervised and unsupervised learning methods can be used to detect geological features from seismic data. As an example, I studied one of the Iranian oilfields to detect the channels and faults.